netron/source/acuity.js

const acuity = {};

acuity.ModelFactory = class {

    async match(context) {
        const obj = await context.peek('json');
        if (obj && obj.MetaData && obj.Layers && Object.keys(obj).length < 256) {
            return context.set('acuity', obj);
        }
        return null;
    }

    async open(context) {
        const metadata = await context.metadata('acuity-metadata.json');
        return new acuity.Model(metadata, context.value);
    }
};

acuity.Model = class {

    constructor(metadata, model, data, quantization) {
        this.name = model.MetaData.Name;
        this.format = `Acuity v${model.MetaData.AcuityVersion}`;
        this.runtime = model.MetaData.Platform;
        this.modules = [new acuity.Graph(metadata, model, data, quantization)];
    }
};

acuity.Graph = class {

    constructor(metadata, model) {
        this.nodes = [];
        this.inputs = [];
        this.outputs = [];
        this.metrics = [];
        const values = new Map();
        const value = (name) => {
            if (!values.has(name)) {
                values.set(name, { name, shape: null });
            }
            return values.get(name);
        };
        let totalFlops = 0;
        for (const [name, layer] of Object.entries(model.Layers)) {
            layer.inputs = layer.inputs.map((input) => {
                return value(input);
            });
            layer.outputs = layer.outputs.map((port) => {
                const output = value(`@${name}:${port}`);
                let shape = null;
                if (layer.op.toLowerCase() === 'input' ||
                    layer.op.toLowerCase() === 'variable') {
                    if (Object.prototype.hasOwnProperty.call(layer.parameters, 'shape') && layer.parameters.shape.length > 0) {
                        shape = layer.parameters.shape;
                    } else if (Object.prototype.hasOwnProperty.call(layer.parameters, 'size') && Object.prototype.hasOwnProperty.call(layer.parameters, 'channels')) {
                        const sizes = layer.parameters.size.split(' ');
                        shape = [0, parseInt(sizes[0], 10), parseInt(sizes[1], 10), layer.parameters.channels];
                    } else if (Object.prototype.hasOwnProperty.call(layer.parameters, 'is_scalar')) {
                        shape = [1];
                    }
                    if (shape && shape.length === 4 && shape[0] === 0) {
                        shape[0] = 1;
                    }
                }
                output.shape = shape;
                return output;
            });
            // Add other layer types (e.g., pooling, batch norm, etc.) as needed.
            if (layer.type === 'Conv2D') {
                const { kernelShape, inputShape, outputShape } = layer;
                const [kH, kW] = kernelShape;
                const [inC] = inputShape;
                const [outC, oH, oW] = outputShape;
                totalFlops += kH * kW * inC * oH * oW * outC;
            } else if (layer.type === 'Dense') {
                const { inputSize, outputSize } = layer;
                totalFlops += inputSize * outputSize;
            }
        }
        this.metrics.push(new acuity.Argument('flops', totalFlops));
        acuity.Inference.infer(model.Layers);
        for (const [name, obj] of values) {
            const type = new acuity.TensorType(null, new acuity.TensorShape(obj.shape));
            const value = new acuity.Value(name, type, null, null);
            values.set(name, value);
        }
        for (const [name, layer] of Object.entries(model.Layers)) {
            switch (layer.op.toLowerCase()) {
                case 'input': {
                    const value = values.get(layer.outputs[0].name);
                    const argument = new acuity.Argument(name, [value]);
                    this.inputs.push(argument);
                    break;
                }
                case 'output': {
                    const value = values.get(layer.inputs[0].name);
                    const argument = new acuity.Argument(name, [value]);
                    this.outputs.push(argument);
                    break;
                }
                default: {
                    const node = new acuity.Node(metadata, name, layer, values);
                    this.nodes.push(node);
                    break;
                }
            }
        }
    }
};

acuity.Node = class {

    constructor(metadata, name, layer, values) {
        const op = layer.op;
        this.name = name;
        this.type = metadata.type(op) || { name: op };
        this.inputs = [];
        this.outputs = [];
        this.attributes = [];
        if (this.type) {
            if (layer.parameters) {
                for (const [name, value] of Object.entries(layer.parameters)) {
                    const meta = metadata.attribute(op, name);
                    const type = meta && meta.type ? meta.type : null;
                    const visible = meta && meta.default !== undefined && meta.default === value ? false : true;
                    const attribute = new acuity.Argument(name, value, type, visible);
                    this.attributes.push(attribute);
                }
            }
        }
        for (let i = 0; i < layer.inputs.length; i++) {
            const input = layer.inputs[i];
            const value = values.get(input.name);
            const name = this.type && this.type.inputs && i < this.type.inputs.length ? this.type.inputs[i].name : `input${i}`;
            const argument = new acuity.Argument(name, [value]);
            this.inputs.push(argument);
        }

        if (this.type && this.type.constants) {
            for (const constant of this.type.constants) {
                // const name = "@" + this.name + ":" + constant.name;
                const type = new acuity.TensorType(null, new acuity.TensorShape(null));
                const value = new acuity.Value('', type, null, new acuity.Tensor(type));
                const argument = new acuity.Argument(constant.name, [value]);
                this.inputs.push(argument);
            }
        }

        for (let i = 0; i < layer.outputs.length; i++) {
            const output = layer.outputs[i];
            const value = values.get(output.name);
            const name = this.type && this.type.outputs && i < this.type.outputs.length ? this.type.outputs[i].name : `output${i}`;
            const argument = new acuity.Argument(name, [value]);
            this.outputs.push(argument);
        }
    }
};

acuity.Argument = class {

    constructor(name, value, type = null, visible = true) {
        this.name = name;
        this.value = value;
        this.type = type;
        this.visible = visible;
    }
};

acuity.Value = class {

    constructor(name, type = null, quantization = null, initializer = null) {
        if (typeof name !== 'string') {
            throw new acuity.Error(`Invalid value identifier '${JSON.stringify(name)}'.`);
        }
        this.name = name;
        this.type = type;
        this.quantization = quantization;
        this.initializer = initializer;
    }
};

acuity.TensorType = class {

    constructor(dataType, shape) {
        this.dataType = dataType || '?';
        this.shape = shape;
    }

    toString() {
        return (this.dataType || '?') + this.shape.toString();
    }
};

acuity.TensorShape = class {

    constructor(dimensions) {
        this.dimensions = Array.isArray(dimensions) && dimensions.length === 1 && dimensions[0] === 0 ? [] : dimensions;
    }

    toString() {
        if (!Array.isArray(this.dimensions) || this.dimensions.length === 0 || (this.dimensions.length === 1 && this.dimensions[0] === 0)) {
            return '';
        }
        return `[${this.dimensions.map((dimension) => dimension ? dimension.toString() : '?').join(',')}]`;
    }
};

acuity.Tensor = class {

    constructor(type) {
        this.type = type;
        this.Category = 'Constant';
    }
};

acuity.Inference = class {

    static infer(layers) {
        const outputs = new Map();
        const outputLayers = [];
        for (const [, layer] of Object.entries(layers)) {
            if (layer.op.toLowerCase() === 'output') {
                outputLayers.push(layer);
            }
            for (const output of layer.outputs) {
                outputs.set(output.name, layer);
            }
        }
        const broadcasts = new Set([
            'add', 'equal', 'fllor_mod', 'floor_div', 'greater', 'greater_equal', 'less', 'less_equal',
            'logical_and', 'logical_or', 'minimum', 'multiply', 'not_equal', 'pow', 'real_div',
            'squared_difference', 'subtract', 'divide', 'addn', 'Divide', 'bitwise_and', 'bitwise_or',
            'bitwise_xor', 'average', 'logical_not', 'logical_xor'
        ]);
        const passthroughs = new Set([
            'LocalResponseNormalization', 'a_times_b_plus_c', 'abs', 'batchnorm_single', 'batchnormalize',
            'cast', 'cast', 'clipbyvalue', 'dequantize', 'dtype_converter', 'elu', 'exp', 'floor',
            'groupnormalize', 'hard_sigmoid', 'hard_swish', 'instancenormalize', 'l2normalize', 'l2normalizescale',
            'layernormalize', 'leakyrelu', 'log', 'log_softmax', 'mish', 'neg', 'norm_with_channel_mean',
            'norm_with_min_max', 'norm_with_scale', 'pow', 'prelu', 'quantize', 'relu', 'relu_keras',
            'relun', 'reverse', 'round', 'rsqrt', 'sigmoid', 'sin', 'softmax', 'softrelu', 'sqrt', 'square', 'tanh',
            'swish', 'gelu', 'dropout', 'eltwise', 'cos', 'l1_layernormalize', 'inverse_sigmoid', 'selu', 'mod',
            'mish', 'minimum_with_clip', 'celu', 'cumsum', 'dft', 'dropout2', 'erf', 'noop', 'squashing', 'tan', 'ceil',
            'atan', 'atan2', 'atanh', 'alpha_dropout', 'acosh', 'rmsnormalize', 'sign'
        ]);
        const reduces = new Set([
            'reduceany', 'reducemax', 'reducemean', 'reducemin', 'reduceprod', 'reducesum'
        ]);
        const poolings = new Set([
            'pooling', 'l2pooling'
        ]);
        const operators = new Map();
        operators.set('broadcast', ([a, b]) => {
            const longer = a.length >= b.length ? a.slice() : b.slice();
            const shorter = a.length < b.length ? a.slice() : b.slice();
            const remain = longer.length - shorter.length;
            for (let i = 0; i < remain; i++) {
                shorter.splice(0, 0, 1);
            }
            for (let i = 0; i < longer.length; i++) {
                longer[i] = longer[i] > shorter[i] ? longer[i] : shorter[i];
            }
            return [longer];
        });
        operators.set('concat', (inputs, params) => {
            const outputShape = inputs[0].slice();
            outputShape[params.dim] = 0;
            for (const shape of inputs) {
                outputShape[params.dim] += shape[params.dim];
            }
            return [outputShape];
        });
        operators.set('conv1d', (inputs, params) => {
            if (params.padding === 'VALID') {
                const out_h = ~~((inputs[0][1] + params.stride - params.ksize) / params.stride);
                return [[inputs[0][0], out_h, params.weights]];
            } else if (params.padding === 'SAME') {
                const out_h = ~~((inputs[0][1] + params.stride - 1) / params.stride);
                return [[inputs[0][0], out_h, params.weights]];
            }
            return null;
        });
        operators.set('convolution', (inputs, params) => {
            if (params.padding === 'VALID') {
                const out_h = Math.floor((inputs[0][1] + params.stride_h + 2 * params.pad_h - params.ksize_h) / params.stride_h);
                const out_w = Math.floor((inputs[0][2] + params.stride_w + 2 * params.pad_w - params.ksize_w) / params.stride_w);
                return [[inputs[0][0], out_h, out_w, params.weights]];
            } else if (params.padding === 'SAME') {
                const out_h = Math.floor((inputs[0][1] + params.stride_h - 1) / params.stride_h);
                const out_w = Math.floor((inputs[0][2] + params.stride_w - 1) / params.stride_w);
                return [[inputs[0][0], out_h, out_w, params.weights]];
            }
            return null;
        });
        operators.set('depthwise_conv1d', (inputs, params) => {
            if (params.padding === 'VALID') {
                const out_h = ~~((inputs[0][1] + params.stride + params.pad[0] + params.pad[1] - params.ksize) / params.stride);
                return [[inputs[0][0], out_h, inputs[0][2] * params.multiplier]];
            } else if (params.padding === 'SAME') {
                const out_h = ~~((inputs[0][1] + params.stride - 1) / params.stride);
                return [[inputs[0][0], out_h, inputs[0][2] * params.multiplier]];
            }
            return null;
        });
        operators.set('depthwise_convolution', (inputs, params) => {
            if (params.padding === 'VALID') {
                const out_h = ~~((inputs[0][1] + params.stride_h + params.pad[0] + params.pad[1] - params.ksize_h) / params.stride_h);
                const out_w = ~~((inputs[0][2] + params.stride_w + params.pad[2] + params.pad[3] - params.ksize_w) / params.stride_w);
                return [[inputs[0][0], out_h, out_w, inputs[0][3] * params.multiplier]];
            } else if (params.padding === 'SAME') {
                const out_h = ~~((inputs[0][1] + params.stride_h - 1) / params.stride_h);
                const out_w = ~~((inputs[0][2] + params.stride_w - 1) / params.stride_w);
                return [[inputs[0][0], out_h, out_w, inputs[0][3] * params.multiplier]];
            }
            return null;
        });
        operators.set('deconvolution', (inputs, params) => {
            return [params.output_shape.map((item, index) => item === 0 ? inputs[0][index] : item)];
        });
        operators.set('deconvolution1d', (inputs, params) => {
            return [params.output_shape.map((item, index) => item === 0 ? inputs[0][index] : item)];
        });
        operators.set('fullconnect', (inputs, params) => {
            return [inputs[0].slice(0, params.axis).concat([params.weights])];
        });
        operators.set('gather', (inputs, params) => {
            const prefix = inputs[1].slice();
            const suffix = inputs[0].slice(params.axis + 1);
            return [prefix.concat(suffix)];
        });
        operators.set('lstm', (inputs, params) => {
            const [input] = inputs;
            const [a, b] = input;
            let batch = a;
            const output = params.num_proj === null ? params.weights : params.num_proj;
            if (params.time_major) {
                batch = b;
            }
            const newShape = params.return_sequences ? [a, b, output] : [batch, output];
            return [newShape, [batch, output], [batch, params.weights]];
        });
        operators.set('matmul', ([a, b], params) => {
            let newShape = a.slice(0, -2);
            if (params.transpose_a) {
                newShape = newShape.concat(a.slice(-1));
            } else {
                newShape = newShape.concat(a.slice(-2, -1));
            }
            if (params.transpose_b) {
                newShape = newShape.concat(b.slice(-2, -1));
            } else {
                newShape = newShape.concat(b.slice(-1));
            }
            return [newShape];
        });
        operators.set('pad', (inputs, params) => {
            return [inputs[0].map((item, index) => item + params.padding_value[index][0] + params.padding_value[index][1])];
        });
        operators.set('permute', (inputs, params) => {
            return [inputs[0].map((item, index) => inputs[0][params.perm[index]])];
        });
        operators.set('pooling', (inputs, params) => {
            if (params.padding === 'VALID') {
                const out_h = ~~((inputs[0][1] + params.stride_h - params.ksize_h) / params.stride_h);
                const out_w = ~~((inputs[0][2] + params.stride_w - params.ksize_w) / params.stride_w);
                return [[inputs[0][0], out_h, out_w, inputs[0][3]]];
            } else if (params.padding === 'SAME') {
                const out_h = ~~((inputs[0][1] + params.stride_h - 1) / params.stride_h);
                const out_w = ~~((inputs[0][2] + params.stride_w - 1) / params.stride_w);
                return [[inputs[0][0], out_h, out_w, inputs[0][3]]];
            }
            return null;
        });
        operators.set('reduce', (inputs, params) => {
            const newShape = inputs[0].slice();
            const axis_list = params.axis_list.map((item) => {
                return item < 0 ? newShape.length + item : item;
            });
            axis_list.sort((a, b) => {
                return b - a;
            });
            axis_list.forEach((i) => {
                newShape[i] = 1;
            });
            if (!params.keep_dims) {
                axis_list.forEach((i) => {
                    newShape.splice(i, 1);
                });
                if (!newShape.length) {
                    newShape.splice(0, 0, 0);
                }
            }
            return [newShape];
        });
        operators.set('repeat', (inputs, params) => {
            const newShape = inputs[0].slice();
            newShape[params.axis] = params.maxlen;
            return [newShape];
        });
        operators.set('reshape', (inputs, params) => {
            const negativeIndexs = [];
            let shape = params.shape;
            if (typeof params.shape === 'string') {
                shape = params.shape.split(/\s+/).map((item) => {
                    return parseInt(item, 10);
                });
            }
            const newShape = shape.map((item, index) => {
                if (item === 0) {
                    return inputs[0][index];
                }
                if (item === -1) {
                    negativeIndexs.push(index);
                    return 1;
                }
                return item;
            });
            if (negativeIndexs.length > 0) {
                newShape[negativeIndexs[0]] = inputs[0].reduce((a, c) => a * c) / newShape.reduce((a, c) => a * c);
            }
            return [newShape];
        });
        operators.set('sequence_mask', (inputs, params) => {
            return [inputs[0].slice().concat([params.maxlen])];
        });
        operators.set('slice', (inputs, params) => {
            return [params.size.map((item, index) => item === -1 ? inputs[0][index] : item)];
        });
        operators.set('squeeze', (inputs, params) => {
            const newShape = inputs[0].slice();
            const axis_list = [...new Set(params.axis_list)].sort((a, b) => b - a);
            for (const item of axis_list) {
                newShape.splice(item, 1);
            }
            return [newShape];
        });
        operators.set('space2depth', (inputs, params) => {
            const h = inputs[0][1] / params.block_size[0];
            const w = inputs[0][2] / params.block_size[1];
            const c = inputs[0][3] * params.block_size[1] * params.block_size[1];
            return [[inputs[0][0], h, w, c]];
        });
        operators.set('depth2space', (inputs, params) => {
            const h = inputs[0][1] * params.block_size;
            const w = inputs[0][2] * params.block_size;
            const c = inputs[0][3] / (params.block_size * params.block_size);
            return [[inputs[0][0], h, w, c]];
        });
        operators.set('upsampling', (inputs, params) => {
            const h = inputs[0][1] * params.factor;
            const w = inputs[0][2] * params.factor;
            return [[inputs[0][0], h, w, inputs[0][3]]];
        });
        operators.set('crop_image', (inputs, params) => {
            return [[inputs[0][0], params.crop_size[0], params.crop_size[1], inputs[0][3]]];
        });
        operators.set('split', (inputs, params) => {
            const sizes = [];
            const slices = params.slices.slice();
            slices.splice(0, 0, 0);
            slices.push(inputs[0][params.dim]);
            slices.reduce((a, b) => {
                sizes.push(b - a);
                return b;
            });
            return sizes.map((item) => {
                const shape = inputs[0].slice();
                shape[params.dim] = item;
                return shape;
            });
        });
        operators.set('stack', (inputs, params) => {
            const newShape = inputs[0].slice();
            if (newShape.length === 1 && newShape[0] === 0) {
                newShape[0] = 1;
            } else {
                newShape.splice(params.axis, 0, inputs.length);
            }
            return [newShape];
        });
        operators.set('stridedslice', (inputs, params) => {
            const input_shape = inputs[0].slice();
            const begin = params.slice_begin.slice();
            const end = params.slice_end.slice();
            if (params.slice_begin_mask > 0) {
                for (let i = 0; i < begin.length; i++) {
                    if ((params.slice_begin_mask >>> i) & 0x1) {
                        begin[i] = -1;
                    }
                }
            }
            if (params.slice_end_mask > 0) {
                for (let i = 0; i < end.length; i++) {
                    if ((params.slice_end_mask >>> i) & 0x1) {
                        end[i] = -1;
                    }
                }
            }
            for (let i = 0; i < begin.length; i++) {
                if (begin[i] === -1) {
                    begin[i] = 0;
                }
            }
            if (inputs[0].length === end.length) {
                for (let i = 0; i < end.length; i++) {
                    if (end[i] === -1 || end[i] > input_shape[i]) {
                        end[i] = input_shape[i];
                    }
                }
            } else if (inputs[0].length < end.length) {
                if (params.slice_new_axis_mask) {
                    const len = (params.slice_new_axis_mask >>> 0).toString(2).length;
                    for (let i = 0; i < len; i++) {
                        if ((params.slice_new_axis_mask >>> i) & 0x1) {
                            input_shape.splice(i, 0, 1);
                        }
                    }
                    for (let i = 0; i < end.length; i++) {
                        if (end[i] === -1) {
                            end[i] = input_shape[i];
                        }
                    }
                }
            }
            let newShape = [];
            for (let i = 0; i < begin.length; i++) {
                newShape = newShape.concat([(end[i] - begin[i]) / params.slice_strides[i]]);
            }
            if (params.slice_shrink_axis_mask) {
                const len = (params.slice_shrink_axis_mask >>> 0).toString(2).length;
                for (let i = 0; i < len; i++) {
                    if ((params.slice_shrink_axis_mask >>> i) & 0x1) {
                        newShape.splice(i, 1);
                    }
                }
            }
            if (params.slice_new_axis_mask) {
                const len = (params.slice_new_axis_mask >>> 0).toString(2).length;
                for (let i = 0; i < len; i++) {
                    if ((params.slice_new_axis_mask >>> i) & 0x1) {
                        if (inputs[0].length === begin.length) {
                            newShape.splice(i, 0, 1);
                        } else if (inputs[0].length < begin.length) {
                            newShape[i] = 1;
                        }
                    }
                }
            }
            return [newShape];
        });
        operators.set('image_resize', (inputs, params) => {
            const newShape = inputs[0].slice();
            newShape[1] = params.new_size[0];
            newShape[2] = params.new_size[1];
            return [newShape];
        });
        operators.set('argmax', (inputs, params) => {
            const newShape = inputs[0].slice();
            if (params.keepdims) {
                newShape[params.axis] = 1;
            } else {
                newShape.splice(params.axis, 1);
                if (!newShape.length) {
                    newShape.splice(0, 0, 0);
                }
            }
            return [newShape];
        });
        operators.set('argmin', operators.get('argmax'));
        /* eslint-disable no-unused-vars */
        operators.set('shapelayer', (inputs, params) => {
            return [[inputs[0].length]];
        });
        operators.set('capsule_norm', (inputs, params) => {
            return [[inputs[0][0], inputs[0][inputs[0].length - 1]]];
        });
        operators.set('size', (inputs, params) => {
            return [[1]];
        });
        /* eslint-enable no-unused-vars */
        operators.set('einsum', ((operators, inputs, params) => {
            const identifyOperation = (inputs, equation) => {
                const identifyFuncs = new Map();
                identifyFuncs.set('matmul', (inputs, equation) => {
                    if (inputs.length !== 2) {
                        return { found: false };
                    }

                    const parts = equation.replace(/\s+/g, '').split(/,|->/);
                    if (parts.length !== 3) {
                        return { found: false };
                    }

                    const [first, second, output] = parts.map((p) => p.split(''));
                    if (!(first.length === output.length || second.length === output.length)) {
                        return { found: false };
                    }

                    let a = first.slice(-2);
                    const b = second.slice(-2);
                    const c = output.slice(-2);
                    let transpose_a = false;
                    let transpose_b = false;
                    if (a[0] === c[0]) {
                        transpose_a = false;
                    } else if (a[1] === c[0]) {
                        transpose_a = true;
                        a = [].concat(a.reverse());
                    } else {
                        return { found: false };
                    }

                    if (a[1] === b[0]) {
                        transpose_b = false;
                    } else if (a[1] === b[1]) {
                        transpose_b = true;
                    } else {
                        return { found: false };
                    }
                    return { found: true, op: 'matmul', params: { transpose_a, transpose_b } };
                });

                /* eslint-disable no-unused-vars */
                for (const [name, func] of identifyFuncs.entries()) {
                    const result = func(inputs, equation);
                    if (result.found) {
                        return result;
                    }
                }
                /* eslint-enable no-unused-vars */
                return { found: false };
            };

            const result = identifyOperation(inputs, params.equation);
            if (result.found) {
                if (operators.has(result.op)) {
                    return operators.get(result.op)(inputs, result.params);
                }
            }
            return [];
        }).bind(undefined, operators));
        const infer = (output) => {
            if (outputs.has(output.name)) {
                let ready = true;
                const layer = outputs.get(output.name);
                for (const input of layer.inputs) {
                    if (input.shape === null) {
                        infer(input);
                        if (input.shape === null) {
                            ready = false;
                            break;
                        }
                    }
                }
                if (ready) {
                    let callback = null;
                    if (operators.has(layer.op)) {
                        callback = operators.get(layer.op);
                    } else if (passthroughs.has(layer.op)) {
                        callback = (inputs) => [inputs[0].slice()];
                    } else if (broadcasts.has(layer.op)) {
                        callback = operators.get('broadcast');
                    } else if (reduces.has(layer.op)) {
                        callback = operators.get('reduce');
                    } else if (poolings.has(layer.op)) {
                        callback = operators.get('pooling');
                    }
                    if (!callback) {
                        callback = () => [];
                    }
                    const parameters = layer.parameters;
                    const inputs = layer.inputs.map((input) => input.shape);
                    const outputs = callback(inputs, parameters);
                    for (let i = 0; i < outputs.length; i++) {
                        if (i < layer.outputs.length) {
                            layer.outputs[i].shape = outputs[i];
                        }
                    }
                }
            }
        };
        for (const layer of outputLayers) {
            for (const output of layer.outputs) {
                infer(output);
            }
        }
    }
};

acuity.Error = class extends Error {

    constructor(message) {
        super(message);
        this.name = 'Error loading Acuity model.';
    }
};

export const ModelFactory = acuity.ModelFactory;